Caste was built, and the build is visible in the vocabulary.

Brāhmaṇa reaches 131.9 in the Vasiṣṭha Dharmasūtra and 54.7 in Manu. It is the most frequent of the four at every stage. Kṣatriya 21.6, vaiśya 43.5, śūdra 68.0 at their peaks.

One caveat: brāhmaṇa also names a genre — the Brāhmaṇa texts. Its high readings in the Kauṣītaki (94.9) and Gopatha (45.4) are inflated by self-reference. The Dharmasūtra and Manu figures are not — those texts are not Brāhmaṇas and have no reason to name the genre.

The word for the group at the top is the most frequent of the four in every law code.

1 · jāti — caste as actually lived — does not exist until the Upaniṣads. Zero in the Rigveda, zero in the Atharvaveda, zero in the Sāmaveda, zero in all three Brāhmaṇas. It appears first in the Bṛhadāraṇyaka and Chāndogya, and reaches 5.0 per 10,000 in Manu. The unit of caste is younger than the philosophy that is supposed to justify it.

2 · śūdra rises 315-fold. One occurrence in the entire Rigveda. In Manu it is one of the most frequent nouns in the text.

3 · mleccha is entirely absent until Manu. Not Rigvedic, not Atharvavedic, not Brāhmaṇic in this sample, not Upaniṣadic. The word for the barbarian is a late arrival.

4 · caṇḍāla appears at 0.1 in the Atharvaveda, 0.9 in the Chāndogya, 3.4 in Manu. The category is invented, then filled in.

śūdra: 1.6 in the Chāndogya → 37.2 in the Āpastamba Dharmasūtra. A twenty-three-fold jump in a single step. vaiśya: 0.4 → 28.6. varṇa: 2.7 → 22.9.

This is not a gradual accumulation across a thousand years. The Vedic and Upaniṣadic material sits flat and low. Then the Dharmasūtras arrive and the vocabulary saturates immediately — and it stays saturated through Manu and the later Smṛtis, seven hundred years on.

Caste vocabulary does not grow into the tradition. It arrives with a genre.

And the genre is law. The texts that carry it are not hymns, not ritual manuals, not philosophy — they are codes of conduct with penalties attached. Vasiṣṭha reaches 68.0 for śūdra, higher than Manu.

Zero occurrences. Not in the Rigveda, not in the Atharvaveda, not in the Sāmaveda, not in any of the three Brāhmaṇas sampled, not in the Bṛhadāraṇyaka, not in the Chāndogya. It appears in Manu.

And it is the entire basis of Arjuna's refusal to fight. Six verses at Bhagavad Gītā 1.39–44 — family destroyed, women corrupted, castes mixed, ancestors fallen — resting on a concept that has no attestation anywhere in the Vedic corpus.

The moral crisis at the opening of the Gītā is built on a word the Vedas never use.

The Gītā sits inside the Mahābhārata, which is not in this sample and should be. But the Vedic silence is complete, and it is the Vedas the Gītā claims continuity with.

166,000 words. Zero jāti. Zero caṇḍāla. Zero dvija. Zero varṇasaṃkara. Zero antyaja.

And varṇa at 0.5 — less than half the Rigveda's rate, in a text roughly the same length.

This is the text that contains the Bṛhadāraṇyaka Upaniṣad as its final section. It is the fullest surviving account of Vedic ritual practice — the sacrifices, the fire altars, the rājasūya, the aśvamedha, described procedure by procedure across fourteen books.

The most detailed record of what Vedic religion actually consisted of contains none of the vocabulary the religion is now identified with.

What it does have is kṣatriya at 2.2 — its highest rate anywhere before the law codes — which fits a text preoccupied with royal consecration. The word is doing work about kingship, not about a birth order.

And dharma at 2.0, lower than the Rigveda's 3.4. The term that reaches 82.5 in Manu is almost absent from the ritual literature it supposedly grows out of.

Each category has more than one word, and using only one hides material.

The untouchability vocabulary was undercounted by a factor of three, because Manu uses five different words: caṇḍāla 24, niṣāda 18, antyaja 14, pukkasa 12, śvapāka 6.

Vedic Saṃhitās: 732,756 words. Brāhmaṇas: 107,492 words. Occurrences of jāti: none.

Everywhere else in Sanskrit literature the word is present. It is in the Upaniṣads, the epics, the Purāṇas, the philosophical corpus, the Buddhist corpus, the law books and the Arthaśāstra. The two sections where it does not occur are the two oldest.

Nearly eighteen million words of Sanskrit contain the word for birth-group. The eight hundred thousand oldest do not.

Śānti and Anuśāsana are the two parvans textual scholarship treats as the largest late didactic accretions — the deathbed instruction of Bhīṣma, a body of teaching attached to the narrative rather than arising from it.

They carry the caste vocabulary and the rest of the epic does not.

Anuśāsana has śūdra at 11.2 — a hundred times the Droṇa parvan's 0.1, and nine times the Bhīṣma parvan that contains the Gītā.

Śānti has dharma at 101.1 — higher than Manu's 82.5. A section of an epic is denser in the word than the law code itself.

The story does not carry the caste material. The instruction attached to the story does.

This is a textual argument made independently of the counts — Sukthankar's Critical Edition and a century of philology identify these parvans as accretions on stylistic and manuscript grounds. The vocabulary distribution agrees with that conclusion without being derived from it.

Rigveda · Paippalāda Atharvaveda · Sāmaveda · Taittirīya Saṃhitā · four Brāhmaṇas including the Śatapatha · the Bṛhadāraṇyaka.

Jāti: zero. Dvija: zero outside a handful in the Rigveda. Caṇḍāla: zero.

Seven hundred thousand words of Vedic text without a single occurrence of the word that names caste as it is actually lived.

The Taittirīya is the Yajurveda of the south — the recension of the Taittirīya śākhā, the dominant Vedic school of Tamil Nadu, Andhra, Karnataka and Kerala. It has all four varna words, at low rates, and none of the caste machinery.

Method note: this text is a padapāṭha — each word separated for recitation — which inflates the token count relative to continuous prose. The rates above are therefore conservative; the true per-word frequencies are higher, and the zeroes are unaffected.

The GRETIL Bṛhadāraṇyaka file is 85,467 words. The Upaniṣad itself is about 17,000. The remainder is Śaṅkara's commentary, written around the eighth century CE — roughly 1,300 years after the text.

So every figure this page reported for "the Upaniṣads" was measuring Śaṅkara.

The fix is in the same source. The Bṛhadāraṇyaka is the fourteenth kāṇḍa of the Śatapatha Brāhmaṇa. Taking that book on its own gives the Upaniṣad with no commentary at all — and confirms it: 106 of the 145 occurrences of Yājñavalkya's name, and all 53 of Gārgī's, are in kāṇḍa 14.

The claim that jāti first appears in the Upaniṣads is wrong. It does not appear in the Bṛhadāraṇyaka at all. It appears in an eighth-century commentary on the Bṛhadāraṇyaka.

The corrected result is stronger and simpler:

Jāti — the endogamous birth-group, caste as actually lived — is absent from the Rigveda, the Atharvaveda, the Sāmaveda, four Brāhmaṇas, and the Bṛhadāraṇyaka Upaniṣad. Roughly half a million words. Zero occurrences.

It enters with the law codes. Everything earlier is either silence or a commentator writing a thousand years later.

And the same correction applies to the untouchability vocabulary. Last week's count found pukkasa and śvapāka in the Bṛhadāraṇyaka. They are Śaṅkara's words, not the Upaniṣad's.

The Chāndogya file has the same problem and no clean equivalent is available here. Its caṇḍāla at 5.10.7 is genuinely in the text — that passage is well attested — but its jāti figure should be treated as unverified until the commentary can be stripped.

A law code about social duties will mention social categories more than a ritual manual does — regardless of when it was written. The Dharmasūtras are about who may do what. The Brāhmaṇas are about how to build a fire altar. Some of the rise measures subject matter, not time.

Three things limit that objection without removing it:

1 · The Brāhmaṇas do discuss social matters — who may officiate, who may attend, who may receive. The Śatapatha spends 166,000 words on ritual and still records zero jāti, zero dvija, zero untouchability terms. A ritual text with strict eligibility rules and no vocabulary for birth-rank is informative.

2 · The Dharmasūtras are short. Āpastamba is 3,498 words; Gautama 4,636. Small denominators make rates volatile — a handful of occurrences produces a large figure. Manu, at 70,538 words, is the stable anchor and it still shows the pattern.

3 · And the genre effect can be isolated exactly. The Bṛhadāraṇyaka is the fourteenth book of the Śatapatha — same school, same recension, same manuscript. Between its ritual books and its philosophical one, dharma rises six-fold, 1.3 to 8.2 — while jāti, dvija and the untouchability vocabulary stay at zero in both.

Genre moves dharma. It does not move the caste vocabulary, because the caste vocabulary is not there to move.